tag:blogger.com,1999:blog-7277449027963623452.post9005392370741250485..comments2023-12-07T05:43:10.401-08:00Comments on Ray Tracey's blog: Incoherent ray tracing without acceleration structuresSam Laperehttp://www.blogger.com/profile/05688552048697970050noreply@blogger.comBlogger9125tag:blogger.com,1999:blog-7277449027963623452.post-91441207996385892272012-10-14T02:07:55.289-07:002012-10-14T02:07:55.289-07:00I don't know the exact number but I think it d...I don't know the exact number but I think it does help a lot. Just think about all the coalesced memory accesses that wont happen otherwise, as well as the improved SIMD-efficiency. You can enable/disable it in their demo program. I may test it later :) This is from the README:<br /><br />"The sorting of is performed by the CPU and is a quite heavy operation, taking roughly one second per batch. It is disabled in the interactive mode."<br /><br />They wouldn't do that if there wasn't a good performance boost :)Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-12438236542835530682012-10-13T18:17:05.957-07:002012-10-13T18:17:05.957-07:00That's interesting. Do you happen to know how ...That's interesting. Do you happen to know how much speed gain there is on average when sorting rays?Sam Laperehttps://www.blogger.com/profile/05688552048697970050noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-28452478780633177322012-10-13T15:51:10.619-07:002012-10-13T15:51:10.619-07:00Note that Aila and Laine sorts all rays on the CPU...Note that Aila and Laine sorts all rays on the CPU in order to increase coherence. This is not included in their Mrays/sec (I believe).Anonymousnoreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-68215823724839470872012-04-24T05:18:50.919-07:002012-04-24T05:18:50.919-07:00cool :)cool :)Sam Laperehttps://www.blogger.com/profile/05688552048697970050noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-34749834686771317412012-04-24T05:17:31.172-07:002012-04-24T05:17:31.172-07:00I will certainly think about it. :)I will certainly think about it. :)Attila T. Áfrahttps://www.blogger.com/profile/10083312082240865646noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-50604731084991938762012-04-24T05:11:28.621-07:002012-04-24T05:11:28.621-07:00Thanks Atilla. Do you have any plans to port your ...Thanks Atilla. Do you have any plans to port your algorithm to the GPU or is it not that well suited for parallelization (as mentioned at the end of the paper)?Sam Laperehttps://www.blogger.com/profile/05688552048697970050noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-87833908511457278682012-04-24T04:28:17.369-07:002012-04-24T04:28:17.369-07:00I think he refers to the GPU ray tracer by Aila an...I think he refers to the GPU ray tracer by Aila and Laine:<br />http://www.tml.tkk.fi/~timo/HPG2009/index.html<br /><br />Those results are indeed very impressive. However, the gap is much narrower on the CPU, and I think that CPU ray tracing is still very important. I've compared the results to a highly optimized MBVH ray tracer, which is one of the fastest on the CPU (the AVX version is faster than Embree). The difference is not that big on one thread, but on multiple threads the bandwidth limitation sadly kicks in. In the worst case, it's about 4x slower.Attila T. Áfrahttps://www.blogger.com/profile/10083312082240865646noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-57103502416087062812012-04-24T03:40:26.104-07:002012-04-24T03:40:26.104-07:00What star methods are you referring to and are the...What star methods are you referring to and are they cpu or gpu based?Sam Laperehttps://www.blogger.com/profile/05688552048697970050noreply@blogger.comtag:blogger.com,1999:blog-7277449027963623452.post-37940493253321084222012-04-24T03:37:59.171-07:002012-04-24T03:37:59.171-07:00nice academic work but i must observe that the res...nice academic work but i must observe that the results (mrays/s) are a factor of 15x behind star methods (accl. structur rebuild included).<br /><br />these scenes can be rebuild in a handful of ms + fast rt incoherent rays. the boeing would be a nice example but the io pressure will limit the approach again to 1-2 fps only.Anonymousnoreply@blogger.com